Explanation

This analysis explores the agreement of two labelers on whether calls both identified as focal on two recordings are the same/different and on who the focal caller was. For now, it uses the output from identify_possible_misidentified_focal_calls.R which was run independently by both Vlad (labeler 1) and Baptiste (labeler 2) to visually inspect and hear calls that were identified on two collar recordings as being “focal” and overlapped in time (see identify_possible_misidentified_focal_calls.R for further info). Here we will take a look at the output to assess the level of agreement between the two labelers, as well as how many “matches” were found and how these things are distributed across call types.

## Loading required package: spam
## Loading required package: dotCall64
## Loading required package: grid
## Spam version 2.5-1 (2019-12-12) is loaded.
## Type 'help( Spam)' or 'demo( spam)' for a short introduction 
## and overview of this package.
## Help for individual functions is also obtained by adding the
## suffix '.spam' to the function name, e.g. 'help( chol.spam)'.
## 
## Attaching package: 'spam'
## The following objects are masked from 'package:base':
## 
##     backsolve, forwardsolve
## Loading required package: maps
## See https://github.com/NCAR/Fields for
##  an extensive vignette, other supplements and source code
## Loading required package: viridisLite

Note about the input data

Note that due to some slight differences in which version of the call labels were used, a few calls are not the same between the labels generated by Vlad and those later generated by Baptiste. In more detail, there were 2942 matches that both labelers looked at, 60 unique to labeler 1 (vlad), and 61 unique to labeler 2 (bapt). We could explore this in more detail later if needed. This also meant that I had to create a new unique identifier for the matches, because the unique identifiers from Baptiste’s and Vlad’s files did not match. I constructed the ‘match.id’ from the two file names, the two recording start times, and the two durations. The following analysis only looks at the matches looked at by both Baptiste and Vlad.

Some basic info

First, let’s get some info on how many calls of each type we have in our ‘matched calls’ dataset. Here is a table of the call types, sorted by how common they are (note becuase this uses both sides of the match, the numbers are essentially duplicated).

Var1 Freq
s 3201
cc 1003
chat 459
agg 360
soc 329
al 160
cc+agg 59
ukn 31
s+soc 27
s+s 26
mov+s 22
s+cc 22
cc+soc 17
mov 17
x 16
mo 15
mo+s 13
s+mo 13
alarm 9
ld 8
mo+ld 8
s+mov 5
soc+agg 5
unk 5
al? 4
cc+ 4
s+al 4
s* 3
s+c 3
soc+s 3
beep 2
cc+ld 2
cc+s 2
ld+cc 2
ld+mo 2
mov+cc 2
mov+ld 2
s+ld 2
sc 2
sn 2
soc* 2
soc+cc 2
soc+mo 2
# 1
agg+soc 1
eating 1
f 1
Marker 1
s+alarm 1
seq 1

Agreement

Let’s look at the overall level of agreement between Baptiste and Vlad on whether two calls were the same or not. We’ll use an “agreement matrix” to quantify this. Rows represent Baptiste’s label and columns given Vlad’s label. The possible labels are unknown, yes, no, and these are row/column 1, 2, and 3 respectively.

Agreement about who the focal is

Now let’s look at only the matches where both labelers agreed that the call was the same (i.e. answered ‘yes’). Did they agree on who the caller was (individual 1 or 2)? First, let’s just see how many calls we are talking about, and what calls they are.

Var1 Freq
s 527
soc 144
cc 72
agg 52
chat 37
s+soc 13
ukn 12
cc+agg 10
al 5
cc+soc 5
mo+s 3
s+mo 3
s+s 3
sc 2
soc+mo 2
agg+soc 1
ld 1
mov 1
s* 1
s+al 1
s+alarm 1
s+c 1
s+cc 1
sn 1
soc+agg 1
soc+cc 1
soc+s 1

Now let’s look at the agreement in the same way as before. We’ll ignore a few unknowns and mistypings in Baptiste’s labels (there were only a couple of them).

The two labelers agree about the focal individual 87.4 % of the time

Agreement on focal by call type

## Warning in grep("cc", compare.yeses$type.a) | grep("cc", compare.yeses$type.b):
## longer object length is not a multiple of shorter object length

When it comes to close calls in particular (broadly defined), the labelers agree 87 % of the time.

How far apart (in time) are the calls labeled as the same call?

As a check of both our accuracy in synching, and whether the calls labeled as “the same” are likely to actually be the same, let’s have a look at how far apart the calls meant to be “the same” are in time. Here is a histogram of the time differences (from the synched GPS times) between calls that have been identified as “the same”e

The onsets for the vast majority of calls are less then 50 ms apart. Very promising! For comparison, here is the histogram for calls where both labelers agreed they were NOT the same.

We can also get a sense of how accurate our labelers are in measuring the duration of calls (though this will likely be an underestimate because calls that actually originate from a meerkat who is not wearing the collar might be harder to measure accurately). Here is a histogram of the differences in estiamted duration (in msec) from calls that were on two recordings and identified as the same call. I’ve plotted them vs. the average duration between the two labeled calls, since probably this also plays a role.

Looks like we are broadly accurate within about 20 msec, most of the time. As expected, we are more accurate on shorter calls.

Distance apart

If we are picking up something real, we’d expect that the calls identified as the same should come from meerkats that are nearby to one another. Is this true? Let’s look at the distribution of distance apart of the two meerkats for calls identified as the same.

We’ll also compare to the distances of the ones where calls were not identified as the same.

Looks like the shorter distances are over-represented in the first plot (where calls were the same). Makes sense. Let’s see how much the shorter distances are over-represented there.

The misidentified focal calls more often come form meerkats that are less than ~3m apart. These distances are over-represented amongst the instances where the calls are the same. Distances above ~3m are underrepresented, and this underrepresentation increases with distance. This pattern seems about as expected.

Discrepancy between files used by Vlad and Baptiste

Earlier on, I noted that the match tables used by Vlad and Baptiste were not quite the same. This indicates that a few of the calls must be different between those analyzed by Vlad and those by Baptiste, probably because they were corrected in label files in the meantime. Let’s have a closer look at this by comparing the call tables from Baptiste and Vlad.

First off, let’s check if all the calls are unique within a given table. It turns out there are a few duplicates, though interestingly the acoustic measurements are not necessarily the same. Would be worth looking into these. The table of duplicates from Baptiste’s file (the latest one) is below.

entryName t0File duration date t0_idx_dayTimeline t0GPS_UTC tEndGPS_UTC tMidGPS_UTC ind fileName labeller verifier callID callType isCall nonFocal hybrid noisy unsureType unsureFocal rms peak.freq.meanentire. fundamental.meanentire. max.freq.meanentire. entropy.meanentire. hnr.meanentire. T_Jitter_abs.nf T_Jitter_rel.nf T_Jitter_per.nf Shimmer_abs.nf Shimmer_rel.nf Shimmer_per.nf T_Jitter_abs.f T_Jitter_rel.f T_Jitter_per.f Shimmer_abs.f Shimmer_rel.f Shimmer_per.f closest_neigh_dist closest_neigh_id nbr_neigh_below_5m id_neigh_below_5m nbr_neigh_below_10m id_neigh_below_10m color x_emitted y_emitted t0num unique.id unique.string
4262 soc nf $ 01:17:43.015 0.442 20190713 4284 2019-07-13 13:11:22.717 2019-07-13 13:11:23.15 2019-07-13 13:11:22.938 VHMF001 HM_VHMF001_HTB_R20_20190707-20190719_file_7_(2019_07_13-11_44_59)_135944_LL_BA.wav LL BA 20190713_VHMF001_01:17:43.015_0.442_soc nf $ soc 1 1 0 0 0 0 -49.26 620 470 806 0.658 18.77 NA NA NA NA NA NA NA NA NA NA NA NA 1.1827592 VHMF010 2 VHMF010 VHMM016 3 VHMF010 VHMF015 VHMM016 1 NA NA 1563016283 4262 HM_VHMF001_HTB_R20_20190707-20190719_file_7_(2019_07_13-11_44_59)_135944_LL_BA.wav|01:17:43.015|0.442|soc nf $
9774 s 01:53:30.407 0.037 20190713 6315 2019-07-13 13:45:14.032 2019-07-13 13:45:14.069 2019-07-13 13:45:14.050 VHMF015 HM_VHMF015_RTTB_R05_20190707-20190719_file_7_(2019_07_13-11_44_59)_135944_HB_BA.wav HB BA 20190713_VHMF015_01:53:30.407_0.037_s s 1 0 0 0 0 0 -43.87 921 879 1097 0.478 19.58 0.0004150 0.3997706 0.2376147 0.0167516 0.0492282 0.0291260 0.0010261 0.6936926 0.3439069 0.0225377 0.0673981 0.0452496 0.7817032 VHMM007 3 VHMM007 VHMM008 VHMM016 5 VCVM001 VHMM007 VHMM008 VHMF010 VHMM016 4 NA NA 1563018314 9774 HM_VHMF015_RTTB_R05_20190707-20190719_file_7_(2019_07_13-11_44_59)_135944_HB_BA.wav|01:53:30.407|0.037|s
12143 lc * 02:32:39.021 0.105 20190712 8412 2019-07-12 14:20:11.288 2019-07-12 14:20:11.393 2019-07-12 14:20:11.340 VHMF019 HM_VHMF019_MBTB_R25_20190707-20190719_file_6_(2019_07_12-11_44_59)_125944_FG_BA.wav FG BA 20190712_VHMF019_02:32:39.021_0.105_lc * lc 1 0 0 0 0 1 -37.99 854 854 1015 0.363 26.58 NA NA NA NA NA NA NA NA NA NA NA NA NA NA 0 NA 0 NA 0 NA NA 1562934011 12143 HM_VHMF019_MBTB_R25_20190707-20190719_file_6_(2019_07_12-11_44_59)_125944_FG_BA.wav|02:32:39.021|0.105|lc *
14985 s nf 01:26:06.579 0.047 20190716 4746 2019-07-16 13:19:05.081 2019-07-16 13:19:05.128 2019-07-16 13:19:05.104 VHMM007 HM_VHMM007_LSLT_R17_20190707-20190719_file_10_(2019_07_16-11_44_59)_165944_HB_VD.wav HB VD 20190716_VHMM007_01:26:06.579_0.047_s nf s 1 1 0 0 0 0 -27.45 958 958 1140 0.317 24.26 0.0002526 0.4042105 0.2610526 0.0194996 0.0596105 0.0252169 0.0002733 0.3870206 0.2076696 0.0197947 0.0594217 0.0355779 1.7722934 VHMM023 2 VHMM014 VHMM023 3 VHMM014 VHMF015 VHMM023 4 NA NA 1563275945 14985 HM_VHMM007_LSLT_R17_20190707-20190719_file_10_(2019_07_16-11_44_59)_165944_HB_VD.wav|01:26:06.579|0.047|s nf
15116 cc nf 01:41:52.183 0.088 20190716 5690 2019-07-16 13:34:49.072 2019-07-16 13:34:49.161 2019-07-16 13:34:49.117 VHMM007 HM_VHMM007_LSLT_R17_20190707-20190719_file_10_(2019_07_16-11_44_59)_165944_HB_VD.wav HB VD 20190716_VHMM007_01:41:52.183_0.088_cc nf cc 1 1 0 0 0 0 -48.60 554 477 748 0.683 22.87 NA NA NA NA NA NA NA NA NA NA NA NA 3.1696126 VHMM023 1 VHMM023 1 VHMM023 2 NA NA 1563276889 15116 HM_VHMM007_LSLT_R17_20190707-20190719_file_10_(2019_07_16-11_44_59)_165944_HB_VD.wav|01:41:52.183|0.088|cc nf
15118 cc nf 01:41:52.400 0.082 20190716 5690 2019-07-16 13:34:49.28 2019-07-16 13:34:49.371 2019-07-16 13:34:49.330 VHMM007 HM_VHMM007_LSLT_R17_20190707-20190719_file_10_(2019_07_16-11_44_59)_165944_HB_VD.wav HB VD 20190716_VHMM007_01:41:52.400_0.082_cc nf cc 1 1 0 0 0 0 -28.69 697 401 888 0.788 22.40 NA NA NA NA NA NA NA NA NA NA NA NA 3.1696126 VHMM023 1 VHMM023 1 VHMM023 2 NA NA 1563276889 15118 HM_VHMM007_LSLT_R17_20190707-20190719_file_10_(2019_07_16-11_44_59)_165944_HB_VD.wav|01:41:52.400|0.082|cc nf
15396 cc 02:12:42.739 0.117 20190712 7337 2019-07-12 14:02:16.21 2019-07-12 14:02:16.327 2019-07-12 14:02:16.268 VHMM007 HM_VHMM007_LSLT_R17_20190707-20190719_file_6_(2019_07_12-11_44_59)_125944_FG_BA.wav FG BA 20190712_VHMM007_02:12:42.739_0.117_cc cc 1 0 0 0 0 0 -28.61 821 585 987 0.576 21.21 0.0020341 1.1934041 0.7893801 0.0266368 0.0795459 0.0455189 0.0018522 1.1364514 0.7571896 0.0131180 0.0400213 0.0166616 3.5842262 VCVM001 1 VCVM001 2 VCVM001 VHMM008 2 NA NA 1562932936 15396 HM_VHMM007_LSLT_R17_20190707-20190719_file_6_(2019_07_12-11_44_59)_125944_FG_BA.wav|02:12:42.739|0.117|cc
25590 fu ld+cc 01:05:56.178 0.163 20190718 3843 2019-07-18 13:04:02.206 2019-07-18 13:04:02.369 2019-07-18 13:04:02.287 VCVM001 HM_VCVM001_SOUNDFOC_20190718_BA.WAV BA NA 20190718_VCVM001_01:05:56.178_0.163_fu ld+cc ld+cc 1 0 1 0 0 0 -28.07 711 608 1644 0.205 33.15 0.0005814 0.6462758 0.4023493 0.0480980 0.1421180 0.0815341 0.0004047 0.3674956 0.2364139 0.0125498 0.0378620 0.0208409 10.5517174 VHMF022 0 NA 0 NA 3 NA NA 1563447842 25590 HM_VCVM001_SOUNDFOC_20190718_BA.WAV|01:05:56.178|0.163|fu ld+cc
25598 fu cc+ld 01:06:27.769 0.192 20190718 3875 2019-07-18 13:04:33.796 2019-07-18 13:04:33.987 2019-07-18 13:04:33.891 VCVM001 HM_VCVM001_SOUNDFOC_20190718_BA.WAV BA NA 20190718_VCVM001_01:06:27.769_0.192_fu cc+ld cc+ld 1 0 1 0 0 0 -32.56 812 522 1431 0.215 34.69 0.0009725 0.8299825 0.4986836 0.0481395 0.1425733 0.0841951 0.0006431 0.5131342 0.3306223 0.0129626 0.0390188 0.0234304 20.6973993 VHMF022 0 NA 0 NA 3 NA NA 1563447874 25598 HM_VCVM001_SOUNDFOC_20190718_BA.WAV|01:06:27.769|0.192|fu cc+ld

Now, on to the matter of the calls which are not shared between Baptiste’s and Vlad’s versions of the call table. Let’s first get these.

Turns out the differences stem from the following files:

## [1] "HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)_175944_FG_VD.wav" 
## [2] "HM_VHMM007_LSLT_R17_20190707-20190719_file_13_(2019_07_19-11_44_59)_195944_FG_VD.wav"
## [3] "HM_VHMM023_MBLS_R02_20190707-20190719_file_11_(2019_07_17-11_44_59)_175944_FG_VD.wav"
## [1] "HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)_175944_FG_VD.wav" 
## [2] "HM_VHMM007_LSLT_R17_20190707-20190719_file_13_(2019_07_19-11_44_59)_195944_FG_VD.wav"
## [3] "HM_VHMM023_MBLS_R02_20190707-20190719_file_11_(2019_07_17-11_44_59)_175944_FG_VD.wav"

This means that any results from intersections with those files need to be redone by the first labeler (Vlad)! But the rest should be fine.

The match.ids in Baptiste’s matches that need to be also labeled by Vlad are the following:

x
HM_VHMF010_SOUNDFOC_20190713_BA.WAV|HM_VHMM017_RSTB_R23_20190708-20190720_file_7_(2019_07_13-11_44_59)_135944_LL_VD.wav|01:59:09.237|02:05:43.127|0.177|0.093
HM_VCVM001_SOUNDFOC_20190718_BA.WAV|HM_VHMM016_LTTB_R29_20190707-20190719_file_12_(2019_07_18-11_44_59)_185944_HB_VD.wav|01:51:54.110|01:55:57.957|0.207|0.134
HM_VCVM001_SOUNDFOC_20190719_2_BA.WAV|HM_VHMM007_LSLT_R17_20190707-20190719_file_13_(2019_07_19-11_44_59)_195944_FG_VD.wav|00:22:43.076|02:03:07.621|0.15|0.214
HM_VCVM001_SOUNDFOC_20190717_BA.WAV|HM_VHMM023_MBLS_R02_20190707-20190719_file_11_(2019_07_17-11_44_59)_175944_FG_VD.wav|01:40:18.588|01:49:35.816|0.121|0.192
HM_VCVM001_SOUNDFOC_20190717_BA.WAV|HM_VHMM023_MBLS_R02_20190707-20190719_file_11_(2019_07_17-11_44_59)_175944_FG_VD.wav|01:40:55.451|01:50:12.465|0.148|0.142
HM_VCVM001_SOUNDFOC_20190717_BA.WAV|HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)_175944_FG_VD.wav|01:43:21.275|01:47:31.375|0.121|0.102
HM_VCVM001_SOUNDFOC_20190717_BA.WAV|HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)_175944_FG_VD.wav|01:46:03.887|01:50:14.166|0.148|0.097
HM_VCVM001_SOUNDFOC_20190717_BA.WAV|HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)_175944_FG_VD.wav|01:46:28.987|01:50:38.858|0.155|0.086
HM_VCVM001_SOUNDFOC_20190717_BA.WAV|HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)_175944_FG_VD.wav|01:50:10.546|01:54:20.888|0.157|0.112
HM_VCVM001_SOUNDFOC_20190717_BA.WAV|HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)_175944_FG_VD.wav|01:50:43.731|01:54:54.225|0.177|0.117
HM_VCVM001_SOUNDFOC_20190717_BA.WAV|HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)_175944_FG_VD.wav|01:51:39.001|01:55:49.629|0.126|0.115
HM_VCVM001_SOUNDFOC_20190717_BA.WAV|HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)_175944_FG_VD.wav|01:51:49.844|01:56:00.417|0.093|0.093
HM_VCVM001_SOUNDFOC_20190717_BA.WAV|HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)_175944_FG_VD.wav|01:53:03.480|01:57:14.401|0.122|0.078
HM_VCVM001_SOUNDFOC_20190717_BA.WAV|HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)_175944_FG_VD.wav|01:53:53.518|01:58:04.503|0.105|0.084
HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_FG_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:21:44.897|01:26:53.432|0.121|0.071
HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_FG_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:24:55.450|01:30:03.515|0.093|0.183
HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_FG_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:25:54.827|01:31:02.957|0.128|0.144
HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_FG_VD.wav|HM_VHMF022_MBRS_R22_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_LL_VD.wav|01:40:26.288|01:45:58.375|0.099|0.068
HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_FG_VD.wav|HM_VHMF022_MBRS_R22_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_LL_VD.wav|01:43:44.880|01:49:16.535|0.086|0.139
HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_FG_VD.wav|HM_VHMM014_LSTB_R19_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_BE_VD.wav|01:47:41.471|01:53:14.519|0.104|0.126
HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_FG_VD.wav|HM_VHMM014_LSTB_R19_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_BE_VD.wav|01:47:44.128|01:53:17.182|0.086|0.162
HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_FG_VD.wav|HM_VHMF022_MBRS_R22_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_LL_VD.wav|01:49:03.861|01:54:35.790|0.086|0.081
HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_FG_VD.wav|HM_VHMF022_MBRS_R22_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_LL_VD.wav|01:49:07.198|01:54:38.762|0.082|0.142
HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_FG_VD.wav|HM_VHMF022_MBRS_R22_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_LL_VD.wav|01:50:44.975|01:56:16.890|0.099|0.155
HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_FG_VD.wav|HM_VHMF022_MBRS_R22_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_LL_VD.wav|01:50:45.791|01:56:17.432|0.108|0.11
HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_FG_VD.wav|HM_VHMM007_LSLT_R17_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_LL_VD.wav|01:54:00.169|01:58:54.771|0.034|0.08
HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_FG_VD.wav|HM_VHMM014_LSTB_R19_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_BE_VD.wav|01:54:54.225|02:00:27.834|0.117|0.115
HM_VHMF001_HTB_R20_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_FG_VD.wav|HM_VHMM014_LSTB_R19_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_BE_VD.wav|01:55:06.884|02:00:40.195|0.089|0.143
HM_VHMF001_HTB_R20_20190707-20190719_file_13_(2019_07_19-11_44_59)195944_HB_VD.wav|HM_VHMM007_LSLT_R17_20190707-20190719_file_13(2019_07_19-11_44_59)_195944_FG_VD.wav|01:52:29.298|01:58:16.326|0.101|0.134
HM_VHMF001_HTB_R20_20190707-20190719_file_13_(2019_07_19-11_44_59)195944_HB_VD.wav|HM_VHMM007_LSLT_R17_20190707-20190719_file_13(2019_07_19-11_44_59)_195944_FG_VD.wav|01:58:21.827|02:04:08.684|0.104|0.132
HM_VHMF001_HTB_R20_20190707-20190719_file_13_(2019_07_19-11_44_59)195944_HB_VD.wav|HM_VHMM007_LSLT_R17_20190707-20190719_file_13(2019_07_19-11_44_59)_195944_FG_VD.wav|01:58:47.886|02:04:34.619|0.113|0.124
HM_VHMF022_MBRS_R22_20190707-20190719_file_13_(2019_07_19-11_44_59)195944_HB_VD.wav|HM_VHMM007_LSLT_R17_20190707-20190719_file_13(2019_07_19-11_44_59)_195944_FG_VD.wav|02:05:23.249|02:04:37.806|0.094|0.126
HM_VHMM007_LSLT_R17_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_LL_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:41:10.644|01:41:24.430|0.103|0.151
HM_VHMM007_LSLT_R17_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_LL_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:45:00.360|01:45:13.973|0.124|0.112
HM_VHMM007_LSLT_R17_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_LL_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:45:10.984|01:45:24.391|0.11|0.117
HM_VHMM007_LSLT_R17_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_LL_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:45:17.380|01:45:30.888|0.114|0.153
HM_VHMM007_LSLT_R17_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_LL_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:49:24.920|01:49:38.234|0.127|0.16
HM_VHMM007_LSLT_R17_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_LL_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:50:51.641|01:51:04.720|0.12|0.144
HM_VHMM007_LSLT_R17_20190707-20190719_file_13_(2019_07_19-11_44_59)195944_FG_VD.wav|HM_VHMM017_RSTB_R23_20190708-20190720_file_13(2019_07_19-11_44_59)_195944_LL_VD.wav|02:00:17.047|01:59:51.256|0.121|0.169
HM_VHMM007_LSLT_R17_20190707-20190719_file_13_(2019_07_19-11_44_59)195944_FG_VD.wav|HM_VHMM014_LSTB_R19_20190707-20190719_file_13(2019_07_19-11_44_59)_195944_FG_VD.wav|02:02:22.259|02:03:08.048|0.129|0.151
HM_VHMM007_LSLT_R17_20190707-20190719_file_13_(2019_07_19-11_44_59)195944_FG_VD.wav|HM_VHMM014_LSTB_R19_20190707-20190719_file_13(2019_07_19-11_44_59)_195944_FG_VD.wav|02:03:30.524|02:04:16.724|0.047|0.14
HM_VHMM007_LSLT_R17_20190707-20190719_file_13_(2019_07_19-11_44_59)195944_FG_VD.wav|HM_VHMM014_LSTB_R19_20190707-20190719_file_13(2019_07_19-11_44_59)_195944_FG_VD.wav|02:03:44.323|02:04:30.073|0.113|0.184
HM_VHMM007_LSLT_R17_20190707-20190719_file_13_(2019_07_19-11_44_59)195944_FG_VD.wav|HM_VHMM017_RSTB_R23_20190708-20190720_file_13(2019_07_19-11_44_59)_195944_LL_VD.wav|02:04:43.726|02:04:17.656|0.132|0.148
HM_VHMM008_SHTB_R14_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_LL_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:59:12.021|01:58:31.665|0.039|0.11
HM_VHMM008_SHTB_R14_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_LL_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:59:12.021|01:58:31.973|0.039|0.119
HM_VHMM008_SHTB_R14_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_LL_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:59:12.364|01:58:31.973|0.043|0.119
HM_VHMM008_SHTB_R14_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_LL_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:59:12.364|01:58:32.359|0.043|0.124
HM_VHMM014_LSTB_R19_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_BE_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:45:52.061|01:45:26.972|0.066|0.13
HM_VHMM014_LSTB_R19_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_BE_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:46:00.859|01:45:35.465|0.14|0.135
HM_VHMM014_LSTB_R19_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_BE_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:46:09.200|01:45:44.029|0.176|0.144
HM_VHMM014_LSTB_R19_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_BE_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:46:45.665|01:46:20.710|0.131|0.172
HM_VHMM014_LSTB_R19_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_BE_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:50:38.007|01:50:12.465|0.136|0.142
HM_VHMM014_LSTB_R19_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_BE_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:50:42.047|01:50:16.374|0.171|0.146
HM_VHMM014_LSTB_R19_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_BE_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:50:45.058|01:50:19.209|0.158|0.124
HM_VHMM014_LSTB_R19_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_BE_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:51:49.227|01:51:23.318|0.151|0.082
HM_VHMM017_RSTB_R23_20190708-20190720_file_11_(2019_07_17-11_44_59)175944_BE_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:58:41.194|01:59:17.181|0.11|0.112
HM_VHMM021_MBLT_R01_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_HB_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:44:24.660|01:44:41.710|0.03|0.119
HM_VHMM021_MBLT_R01_20190707-20190719_file_11_(2019_07_17-11_44_59)175944_HB_VD.wav|HM_VHMM023_MBLS_R02_20190707-20190719_file_11(2019_07_17-11_44_59)_175944_FG_VD.wav|01:46:07.131|01:46:23.989|0.131|0.052

They are all ones associated with those 3 files that were resynched EXCEPT the first 2 (HM_VHMF010_SOUNDFOC_20190713_BA.WAV|HM_VHMM017_RSTB_R23_20190708-20190720_file_7_(2019_07_13-11_44_59)135944_LL_VD.wav|01:59:09.237|02:05:43.127|0.177|0.093 HM_VCVM001_SOUNDFOC_20190718_BA.WAV|HM_VHMM016_LTTB_R29_20190707-20190719_file_12(2019_07_18-11_44_59)_185944_HB_VD.wav|01:51:54.110|01:55:57.957|0.207|0.134)

I am still quite mystified by this, since the corresponding calls are actually present in both calls tables from the two files, so I don’t see why they were matched in Baptiste’s but not in Vlad’s version! But anyway, these 2 should also be labeled by Vlad as well.

Comparing amplitude to neighboring calls

One way to potentially identify focal vs non-focal calls is to compare the amplitude on the 2 devices. However, is this a reliable way to do things? Let’s start by looking at all the calls in a given file of a given type, and computing the amplitude of ones labeled “focal” and “non-focal”. Are they distinguishable? Let’s use the calls2 table (Baptiste’s later one) for this, because it should be more up to date.

The plots below show (for each file) the distribution of RMS (amplitude, high-pass filtered; 100 Hz Butterworth) across the calls labeled as CC’s (broadly defined, accessed via grepping for ‘cc’). In blue, the calls labeled “focal” and in red, the calls labeled “nonfocal” by the labeler (black = distribution over all calls in that category). Note that data are only included for files with > 10 nonfocal and > 10 focal calls. Numbers of each type are shown as text in the plots. Legend background for focal microphone recordings is gray.

Overall, this is not looking promising. Clearly, the labelers are not paying attention to absolute amplitude alone when assessing whether to label something as focal or non-focal, or alternatively our measurement of RMS is not reliable.

One question is whether this could be being confounded by noise. Let’s make the same comparison but for calls labeled “noisy” vs those not labeled “noisy”. Here, we will show the “noisy” calls as blue and the non-noisy as black.

The measured amplitude also does not seem to relate to the “noisy” label.

What about for the calls labeled as focal or nonfocal by Baptiste and Vlad, during the focal/non-focal comparison stage? Did Vlad and Baptiste label these based on their overall amplitude, as measured by our RMS measure?

Since there was high agreement, let’s just use Baptiste’s labels (matches2) for this.

There seems to be some signal there, with the one labeled as focal more likely to be the one with the larger amplitude. However, about 34% of the time the quieter call was labeled as focal.